We present two novel methods to automatically learn spatio-temporal dependencies of moving agents in complex dynamic scenes. They allow to discover temporal rules, such as the right of way between different lanes or typical traffic light sequences. To extract them, sequences of activities need to be learned. While the first method extracts rules based on a learned topic model, the second model called DDP-HMM jointly learns co-occurring activities and their time dependencies. To this end we employ Dependent Dirichlet Processes to learn an arbitrary number of infinite Hidden Markov Models. In contrast to previous work, we build on state-of-the-art topic models that allow to automatically infer all parameters such as the optimal number of HMMs necessary to explain the rules governing a scene. The models are trained offline by Gibbs Sampling using unlabeled training data.
%0 Conference Paper
%1 KuettelBreitensteinEtAl10CVPR
%A Kuettel, Daniel
%A Breitenstein, Michael D.
%A Van Gool, Luc
%A Ferrari, Vittorio
%B Proceedings of CVPR'10: IEEE Conference on Computer Vision and Pattern Recognition, San Francisco, CA, USA
%D 2010
%K v1205 ieee paper ai processing image video action recognition analysis temporal spatial rules zzz.vitra
%P 1951-1958
%R 10.1109/CVPR.2010.5539869
%T What's going on? Discovering Spatio-Temporal Dependencies in Dynamic Scenes
%X We present two novel methods to automatically learn spatio-temporal dependencies of moving agents in complex dynamic scenes. They allow to discover temporal rules, such as the right of way between different lanes or typical traffic light sequences. To extract them, sequences of activities need to be learned. While the first method extracts rules based on a learned topic model, the second model called DDP-HMM jointly learns co-occurring activities and their time dependencies. To this end we employ Dependent Dirichlet Processes to learn an arbitrary number of infinite Hidden Markov Models. In contrast to previous work, we build on state-of-the-art topic models that allow to automatically infer all parameters such as the optimal number of HMMs necessary to explain the rules governing a scene. The models are trained offline by Gibbs Sampling using unlabeled training data.
@inproceedings{KuettelBreitensteinEtAl10CVPR,
abstract = {We present two novel methods to automatically learn spatio-temporal dependencies of moving agents in complex dynamic scenes. They allow to discover temporal rules, such as the right of way between different lanes or typical traffic light sequences. To extract them, sequences of activities need to be learned. While the first method extracts rules based on a learned topic model, the second model called DDP-HMM jointly learns co-occurring activities and their time dependencies. To this end we employ Dependent Dirichlet Processes to learn an arbitrary number of infinite Hidden Markov Models. In contrast to previous work, we build on state-of-the-art topic models that allow to automatically infer all parameters such as the optimal number of HMMs necessary to explain the rules governing a scene. The models are trained offline by Gibbs Sampling using unlabeled training data.},
added-at = {2012-05-30T10:49:36.000+0200},
author = {Kuettel, Daniel and Breitenstein, Michael D. and Van Gool, Luc and Ferrari, Vittorio},
biburl = {https://www.bibsonomy.org/bibtex/2a10092c12e8bd6fa245748e02a60704c/flint63},
booktitle = {Proceedings of {CVPR'10}: IEEE Conference on Computer Vision and Pattern Recognition, San Francisco, CA, USA},
doi = {10.1109/CVPR.2010.5539869},
file = {IEEE Digital Library:2010/KuettelBreitensteinEtAl10CVPR.pdf:PDF},
groups = {public},
interhash = {ebdf7c44cbc3abf3b6944bd480f4ea40},
intrahash = {a10092c12e8bd6fa245748e02a60704c},
keywords = {v1205 ieee paper ai processing image video action recognition analysis temporal spatial rules zzz.vitra},
pages = {1951-1958},
timestamp = {2018-04-16T11:33:32.000+0200},
title = {What's going on? {Discovering} Spatio-Temporal Dependencies in Dynamic Scenes},
username = {flint63},
year = 2010
}